This reads the AfD's own YouTube channels: what the party publishes, and what people write in the comments underneath. Both are sorted by the same rules, which is what makes the one comparison here possible: the party's voice set against its audience's, on the same videos.
Every item on this page was sorted by one model, working to one written codebook, at one setting. No researcher decided where any individual comment belonged, and that is the point: an instrument applied uniformly to the whole corpus can be wrong, but it cannot be selectively wrong in the direction somebody wanted it to go.
The figures pool every week analysed so far into a single reading rather than splitting them into weekly slices, and each one is published with the range of uncertainty around it drawn in.
The categories are the model's judgement rather than established fact. Read them as a systematic first pass over a large body of text, and follow the links further down to the material itself.
Video titles and what is actually said in them, taken from the AfD's own channels.
…Comments on those same videos. Replies are skipped, because a reply only makes sense next to the comment it answers.
…Both sides are sorted by the same rules, so what the party says can be set beside what its audience says back. Nothing here is compared to any other party.
Every figure above was computed from the videos below and the comments underneath them. Each is named and links to the original, so any number on this page can be traced back to the material it came from.
| Video | Week | Comments analysed | Transcript pieces | Views |
|---|---|---|---|---|
| Loading… | ||||
On the party side we count videos, not sentences, because ten sentences taken from one video are really one source. That is why the ranges on the party's bars are so much wider than the audience's: eleven videos is a small number, and the page would rather show you that than hide it.
The idea for the AfD Monitor began with a simple question: what do the AfD and its supporters actually say when we look closely at their own words?
Much of the public discussion surrounding the Alternative für Deutschland (AfD) approaches the party primarily through the question of whether it represents a threat to German democracy. At the same time, the party has continued to attract substantial public support and win votes. These two realities can easily produce competing narratives: one that sees the AfD and its supporters primarily through the lens of extremism and democratic threat, and another that sees the party as expressing legitimate concerns that established political actors have failed to address.
This project takes a different starting point. Rather than beginning with a judgement about what the AfD represents, it asks what its rhetoric actually says, and how that rhetoric is received by its audience.
The AfD Monitor analyses the AfD's own YouTube content alongside the comments written by viewers beneath those same videos. The aim is to examine four basic questions:
The analysis is informed by my academic research on political rhetoric, threat framing and group identity, but the classification itself is performed by an AI model using a predefined codebook and the same rules across the entire corpus. No individual comments are manually selected or classified according to whether they support or challenge a particular interpretation. The purpose is not to claim that the model is infallible but to apply the same analytical instrument consistently and make its limitations visible.
The project therefore should not be read as an attempt to prove that the AfD is either a threat to democracy or a legitimate corrective to German politics. It is an attempt to make the underlying rhetoric visible before deciding what it means.
The data, methodology and classifications are presented openly so that the findings can be examined, questioned and, where necessary, challenged. The categories are analytical judgements rather than established facts, and the results should be understood as a systematic first pass over a large body of political discourse.
Ultimately, the goal is simple: to understand political movements not only through what others say about them, but through what they say themselves, and what their audiences say back.
I am Talip Alkhayer, a political scientist specialising in political rhetoric, threat framing, group identity and computational approaches to the analysis of political discourse. My academic research combines theories from political science and social psychology with methods from natural language processing and machine learning to examine how political actors construct threats, identities and narratives.
I developed the AfD Monitor as an extension of my broader research interest: using computational and AI-assisted methods to examine contemporary political discourse at a scale that would be difficult to analyse manually.
I am interested not only in what political actors say, but in the patterns that emerge when thousands of individual pieces of discourse are analysed systematically.
This is an ongoing research project, and it is intended to be questioned.
If you notice a problem with the methodology, disagree with a classification, have suggestions for improving the codebook, or have ideas for extending the analysis, I would be very interested to hear from you.
I am also open to conversations with researchers, journalists, data scientists and others interested in political discourse, computational social science and AI-assisted research.
Every figure on this site comes from a computer reading text and sorting it into categories. These are the instructions it was given — the real ones, read straight from the files it uses, not a description written afterwards.